To prevent problems and capitalise on opportunities before they even occur, the research project SPEEDD proposed a methodology, and developed a prototype for proactive event-driven decision-making. We present the application of this methodology to credit card fraud management. The machine learning component of the SPEEDD prototype supports the online construction of fraud patterns, allowing it to efficiently adapt to the continuously changing fraud types. Moreover, the user interface of the prototype enables fraud analysts to make the most out of the results of automation (complex event processing) and thus reach informed decisions. Unlike most academic research on credit card fraud management, the assessment of the prototype (components) is based on representative transaction datasets, allowing for a realistic evaluation.